中文摘要：

图像中的噪声或非规则细节干扰易导致形态学分水岭产生较严重的过分割，为了在消除过分割的同时尽可能保持图像目标边界的准确定位，提出了一种基于面积约束和自适应梯度修正的分水岭图像分割方法。首先对图像进行梯度变换，采用区域面积约束滤除狭小高梯度尖峰对应的噪声和非规则细节；然后建立梯度级与结构元素大小之间的函数关系，并以相对应的结构元素对梯度图像进行粘性形态学（VM）闭运算，消除低梯度噪声及非规则细节，实现梯度图像的自适应修正，由于VM闭运算对梯度图像进行修正时，对目标仅作轻度或不作修正，因而能够最大限度的保持目标轮廓的准确定位，而对噪声和非规则细节则采用较大尺寸的结构元素进行较大幅度修正，从而消除产生过分割的因素；最后对修正图像进行分水岭分割。实验结果表明，本文方法能够在消除过分割的同时，保持目标轮廓的准确定位。

英文摘要：

Morphological watershed segmentation often leads to a serious over-segmentation due to the noises and irregular details within an image. Watershed segmentation based on area constraint and adap- tive gradient modification is proposed to alleviate over-segmentation and boundary bias. Firstly,the origi- nal image is transformed to a morphological gradient image. In gradient relief, the high gradient ampli- tudes with small area are usually corresponding to salt-like noises or bright regular details, which can be removed by the area constraint. Secondly,the function between gradient levels and morphologica! struc- ture element is established,and the viscous morphological closing operator is utilized to modify the relief of gradient image with different sizes of structuring elements. The effects of viscous closing applied to objects and noise are different, for the obieets regions, the sizes of structuring elements are smaller, which means they are light or less modification, while for noise or details, the larger size structuring elements are employed to modify them heavily. By such an adaptive modification, most irregular local minimums corresponding to the low amplitudes in gradient image caused by details and noise will be removed,while positions of target boundaries have less change. Finally, standard watershed transform is employed to im- plement segmentation. Experiments show that this method can eliminate over-segmentation effectively while preserve the location of object contours.